CATALOG REVIEW · NOVEMBER 10, 2026

Compare the models.

Choose up to three. Start with the task and access you need, then compare capability and cost.

API prices are separate from app subscriptions. Provider facts and benchmark scores have separate check dates below. All included index scores use Artificial Analysis v5.0.1; versions before 5.0 use different tests. Check the source links, or see the latest benchmark results.

Where to start

Based on the dated API rates and Artificial Analysis v5.0.1 scores below. Index points are not a guarantee on your task; test on work you can check.

22 models · 0/3 selected

Anthropic

Claude Opus 5.5

General & coding

Anthropic’s recommended starting point for demanding coding, agent and analysis work. It scores below Fable 5.2 and Gemini 4 Carbon on the index; compare it with Sonnet on work you can check before paying twice the token rate.

Access
Claude Pro and above; Claude API, Bedrock, Google Cloud and Foundry
API / 1M tokens
$4.00 in / $20.00 out
AA index v5.0.1 · max
59 Nov 10 reference
Context, pricing & sources

1M tokens; 128K output. Standard API rates. Batch requests are half price and cached reads cost less; fast mode and long-context pricing differ.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Claude Opus 5.5 (max); checked 2026-11-10.

Platform overview →
Anthropic

Claude Fable 5.2

New
General & coding

Released November 9 for long-horizon reasoning and agent work, and now the top score on the Artificial Analysis index. It costs twice Opus 5.5 per token, and Anthropic still recommends starting with Opus. Mythos 5.2 is the same model without Fable’s cyber and biology safeguards, for approved security and life-sciences organizations only.

Access
Claude Max and Ultra (up to half of weekly limits), Team Premium seats and Enterprise; Pro through usage credits; Claude API, AWS, Google Cloud and Microsoft Azure
API / 1M tokens
$8.00 in / $40.00 out
AA index v5.0.1 · max
65 Nov 10 reference
Context, pricing & sources

1M tokens; 128K output. Standard API rates, down from $10 / $50 for Fable 5.1. Optional fast mode (up to 2.5× faster) costs $16 / $80. Batch requests are half price and cached reads cost less.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Claude Fable 5.2 (max); checked 2026-11-10.

Platform overview →
Anthropic

Claude Sonnet 5.5

General & coding

Half the token price of Opus and positioned for faster responses. Compare effort settings, reasoning-token use and cost per completed task; the index score here is for max effort.

Access
Claude Free and paid plans; Claude API
API / 1M tokens
$2.00 in / $10.00 out
AA index v5.0.1 · max
55 Nov 10 reference
Context, pricing & sources

1M tokens; 128K output. Standard API rates. Batch requests are half price and cached reads cost less; fast mode and long-context pricing differ.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Claude Sonnet 5.5 (max); checked 2026-11-10.

Platform overview →
Anthropic

Claude Haiku 5.5

High volume

Anthropic’s fastest model, aimed at classification, extraction and routing. Useful when latency and volume matter more than the hardest reasoning.

Access
Claude apps and API
API / 1M tokens
$0.10 in / $0.50 out
AA index v5.0.1 · max
36 Nov 10 reference
Context, pricing & sources

1M tokens; 128K output. Lowest listed rate (“from”); check the pricing page for the rate that applies to your usage. Batch and cached reads cost less.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Claude Haiku 5.5 (max); checked 2026-11-10.

Platform overview →
OpenAI

GPT-6 Astra

General & coding

OpenAI’s flagship for difficult reasoning, coding and tool-driven work. Its standard token rates are five times GPT-6.1 Sol’s; compare both at the effort setting your work needs.

Access
API; ChatGPT’s Pro setting (GPT-6 Pro) on Pro, Business and Enterprise; Plus gets it in Work and Codex
API / 1M tokens
$10.00 in / $50.00 out
AA index v5.0.1 · max
57 Nov 10 reference
Context, pricing & sources

1.05M tokens; 128K output. Standard text-token rates; inputs over 272K tokens cost more (Astra $20 / $75). Batch and Flex are half price; tools, caching and faster service have separate rates.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as GPT-6 Astra (max); checked 2026-11-10.

Platform overview →
OpenAI

GPT-6.1 Sol

General & coding

Released at DevDay (Sept 29) as near-Astra performance at a fifth of the price for coding and professional work. A sensible default before paying for Astra.

Access
API and Codex
API / 1M tokens
$2.00 in / $10.00 out
AA index v5.0.1 · max
55 Nov 10 reference
Context, pricing & sources

1.05M tokens. Standard text-token rates; inputs over 272K tokens cost more (Astra $20 / $75). Batch and Flex are half price; tools, caching and faster service have separate rates.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as GPT-6.1 Sol (max); checked 2026-11-10.

Platform overview →
OpenAI

GPT-6.1 Terra

General & coding

A lower-cost option for visual reasoning, computer use and writing. Tool calling uses the Responses API; evaluate screen-based tasks separately from the general Intelligence Index.

Access
OpenAI API; tool calling through Responses
API / 1M tokens
$1.00 in / $5.00 out
AA index v5.0.1 · max
51 Nov 10 reference
Context, pricing & sources

1.05M tokens; 128K output. Standard text-token rates. Above 272K input tokens, the full request costs 2× input/cache rates and 1.5× output. Batch and Flex are half price; tools, cache writes and regional processing may add charges.

Provider reference checked 2026-11-01. Provider documentation ↗ · Benchmark source ↗ · Tested as GPT-6.1 Terra (max); checked 2026-11-10.

Platform overview →
OpenAI

GPT-6 Luna

High volume

OpenAI’s low-cost model for focused tasks, extraction and large request volumes. Not the same as GPT-5.6 Luna, which runs ChatGPT’s Free and Go plans.

Access
API and Codex
API / 1M tokens
$0.10 in / $0.50 out
AA index v5.0.1 · max
28 Nov 10 reference
Context, pricing & sources

1.05M tokens. Standard text-token rates; inputs over 272K tokens cost more (Astra $20 / $75). Batch and Flex are half price; tools, caching and faster service have separate rates.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as GPT-6 Luna (max); checked 2026-11-10.

Platform overview →
Google

Gemini 4 Carbon

Updated
General & coding

Google’s most capable model, generally available since November 10 after an October preview. Google recommends Argon for most coding and knowledge work, and Carbon for the hardest multi-step coding and research problems.

Access
Gemini API paid tier; Gemini app, AI Mode and Antigravity on Google AI Ultra; Gemini for business
API / 1M tokens
$2.00 in / $10.00 out
AA index v5.0.1 · high
60 Nov 10 reference
Context, pricing & sources

1M tokens (shared with output); 65K output by default, up to 1M with long decode. Introductory rates through November 30, 2026; $4 input / $20 output from December 1, 2026. Not on the free API tier; implicit caching only.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Gemini 4 Carbon (high); checked 2026-11-10.

Platform overview →
Google

Gemini 4 Argon

General & coding

Google’s generally available Gemini 4 model since October 13, and the one Google recommends for most software engineering and knowledge work. The newer Carbon is for the hardest problems and needs Google AI Ultra in the app.

Access
Gemini app on Google AI Pro and Ultra; paid Gemini API tier; Antigravity
API / 1M tokens
$2.00 in / $10.00 out
AA index v5.0.1 · high
50 Nov 10 reference
Context, pricing & sources

1M tokens (shared with output). Introductory rates through November 30, 2026; $4 input / $20 output from December 1, 2026. Not available on the free API tier.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Gemini 4 Argon (high); checked 2026-11-10.

Platform overview →
Google

Gemini 3.8 Flash

General & coding

Google’s current Flash model (generally available September 2) for agent workflows and software engineering, with a free API tier. Requests to the deprecated 3.7 Flash are routed here.

Access
Google AI Studio and Gemini API, including the free tier
API / 1M tokens
$0.75 in / $3.75 out
AA index v5.0.1 · high
31 Nov 10 reference
Context, pricing & sources

See model documentation. Promotional standard rates through December 31, 2026; $1.50 input / $7.50 output from January 1, 2027. Grounding and caching cost extra.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Gemini 3.8 Flash (high); checked 2026-11-10.

Platform overview →
Google

Gemini 3.5 Flash-Lite

High volume

A lightweight model for throughput and cost-sensitive tasks such as extraction and tagging. Its index score is well below the Flash model, so validate quality before increasing volume.

Access
Google AI Studio and Gemini API, including the free tier
API / 1M tokens
$0.30 in / $2.50 out
AA index v5.0.1 · as listed (no effort label)
12 Nov 10 reference
Context, pricing & sources

See model documentation. Standard API rates; grounding and cache storage have separate charges.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Gemini 3.5 Flash-Lite; checked 2026-11-10.

Platform overview →
SpaceXSI (xAI)

Grok 4.7

General & coding

Text and image understanding with adjustable reasoning effort, recommended by SpaceXSI for code and chat. The Grok app adds search, voice and other tools beyond the model itself, and Grok models are the only ones in Cursor’s $8 Plus plan.

Access
Grok and Cursor plans; API
API / 1M tokens
$2.00 in / $6.00 out
AA index v5.0.1 · xhigh
35 Nov 10 reference
Context, pricing & sources

500K tokens. Base API rates. A request with 200K or more input tokens is billed at double the rate for all its tokens; cached input is $0.50.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Grok 4.7 (xhigh); checked 2026-11-10.

Platform overview →
Moonshot AI

Kimi K3.1 Pro

Updated
General & coding

Moonshot’s K3.1 model for vision, coding and long-context agent work. Its full weights (a 2.8-trillion-parameter mixture of experts, 104B active) are now on Hugging Face under Moonshot’s own license. The Highspeed variant is a separate API offering.

Access
Kimi API after a successful top-up (minimum $1); open weights on Hugging Face
API / 1M tokens
$2.00 in / $10.00 out
AA index v5.0.1 · as listed (no effort label)
46 Nov 10 reference
Context, pricing & sources

1,048,576 tokens. Standard uncached input/output rates; cached input $0.20. Cache writes are billed separately by retention. Batch costs 60% of standard; Highspeed has separate rates.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Kimi K3.1 Pro; checked 2026-11-10.

Platform overview →
Moonshot AI

Kimi K3

General & coding

The earlier K3 model remains a reference for long-context, coding and vision work. K3.1 Pro is the newer release; no current v5.0 K3 score is included in this comparison.

Access
Kimi app and API; also offered in Perplexity, GitHub Copilot and Ollama Cloud
API / 1M tokens
$3.00 in / $15.00 out
Context, pricing & sources

1M tokens. Uncached input/output rates; cached input is $0.30 per million tokens.

Provider reference checked 2026-10-22. Provider documentation ↗

Platform overview →
DeepSeek

DeepSeek V4.1 Flash 1017

General & coding

The October 17 revision behind the deepseek-flash API alias supports vision, reasoning and tool calls. Its dated v5.0 score belongs to this revision; compare peak and off-peak rates for your workload.

Access
DeepSeek app and API as deepseek-flash
API / 1M tokens
$0.30 in / $1.20 out
AA index v5.0.1 · max
33 Nov 10 reference
Context, pricing & sources

1M tokens; up to 384K output. Peak uncached-input/output rates shown (01:00–04:00 and 06:00–10:00 UTC on weekdays). Off-peak rates are half: $0.15 / $0.60. Web search and fetch tools are billed per request from November 1, 2026.

Provider reference checked 2026-11-01. Provider documentation ↗ · Benchmark source ↗ · Tested as DeepSeek V4.1 Flash 1017 (max); checked 2026-11-10.

Platform overview →
Meta

Muse Wonder

General & coding

Meta’s Wonder model supports agent workflows and is available through the Model API. Compare its task performance with other models at the same $4 / $20 standard token rates.

Access
Meta Model API (preview) and Muse Code; coming to the Muse app
API / 1M tokens
$4.00 in / $20.00 out
AA index v5.0.1 · max
44 Nov 10 reference
Context, pricing & sources

1,048,576 tokens. Standard rates; cached input is $0.20. Web search grounding adds $2.50 per 1,000 queries. No lower-priced Contributor option.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Muse Wonder (max); checked 2026-11-10.

Platform overview →
Meta

Muse Wonder Full

General & coding

The stronger Wonder variant has a separate benchmark result from the API model. It is offered through Muse Code subscriptions, with no per-token API price included here.

Access
Muse Code Power Usage ($50/mo) and Maximum Usage ($200/mo) only
API / 1M tokens
See provider pricing
AA index v5.0.1 · max
50 Nov 10 reference
Context, pricing & sources

See model documentation. Subscription only; no per-token price.

Provider reference checked 2026-10-22. Provider documentation ↗ · Benchmark source ↗ · Tested as Muse Wonder Full (max); checked 2026-11-10.

Platform overview →
Meta

Muse Spark 1.3

High volume

The model behind the Muse app and Meta’s recommended Spark version for new work. Its Contributor version costs a fraction as much, but Meta trains on what you send it.

Access
Muse app, Muse Code and Meta Model API
API / 1M tokens
$1.25 in / $4.25 out
Context, pricing & sources

1,048,576 tokens. Standard rates (not used for training); cached input $0.15. The Contributor version is $0.10 / $0.20, and Meta uses its prompts and outputs to improve its products.

Provider reference checked 2026-10-22. Provider documentation ↗

Platform overview →
Meta

Muse Ember

New
High volume

Released November 9 as Meta’s low-cost execution model for sub-agent tasks, tool calls, routing and large batches. Meta suggests pairing it with Spark or Wonder: the larger model plans, Ember handles the volume.

Access
Meta Model API (Standard mode only) and every Muse Code plan
API / 1M tokens
$0.10 in / $0.50 out
AA index v5.0.1 · max
36 Nov 10 reference
Context, pricing & sources

1,048,576 tokens; 524,288 output. Rates for requests up to 200K input tokens; a larger request is billed at $0.30 / $1.50 in full. Cached input $0.004. No Contributor discount.

Provider reference checked 2026-11-10. Provider documentation ↗ · Benchmark source ↗ · Tested as Muse Ember (max); checked 2026-11-10.

Platform overview →
OpenAI

GPT Image 2.6 Sunburst

New
Image generation

Released November 4 with Images 2.6, for detail and consistency across edits at medium or high quality. Early Arena votes place it second, behind GPT Image 2.5 Sunburst; test text rendering and composition on your own brief.

Access
OpenAI API (Image API and Responses); Images 2.6 also runs in ChatGPT and Codex on every plan
API / 1M tokens
See provider pricing
Context, pricing & sources

Image workflow. Image tokens: $8 input / $30 output per million; text input $5. Per-image cost depends on size and quality; OpenAI’s image guide has a calculator.

Provider reference checked 2026-11-10. Provider documentation ↗

Platform overview →
OpenAI

GPT Image 2.6 Flare

New
Image generation

The faster Images 2.6 variant: as little as six seconds at low quality, with better layouts for posters and product images. Same listed rates as Sunburst.

Access
OpenAI API (Image API and Responses)
API / 1M tokens
See provider pricing
Context, pricing & sources

Image workflow. Same token rates as Sunburst. Per-image cost depends on size and quality.

Provider reference checked 2026-11-10. Provider documentation ↗

Platform overview →

Model, app or agent?

Model

The engine that generates an answer. A model’s context window and reasoning settings affect what it can process.

App & tools

Search, files, connectors and editing features around that engine. Two apps using the same model can behave very differently.

Agent

A workflow that can take multiple steps and use tools. Finished-task quality depends on instructions, permissions and checks as well as the model.

Compare platform features →